Susie Grigson at Flinders University
Image: Susie GrigsonI am a visiting PhD student from Flinders University in Australia.
Approximately 30% of bacterial and 65% of viral protein sequences cannot be assigned a known biological function. To gain a better understanding of these microbes, my PhD research aims to develop new computational techniques and tools to decrease this ever-increasing sequence function gap. I use machine learning, sequence embeddings and computed protein structures combined with a range of genomic properties including gene arrangements to build software which allows researchers to annotate unknown microbial genes.
I have investigated how amino acid sequence embeddings may be used to study functional hierarchies for describing bacterial protein functions. Using a Protvec sequence embedding, a technique used to represent amino acid sequences in n-dimensional space, I demonstrated that the bacterial carbohydrate metabolism class within the SEED annotation system contains 48 clusters of embedded sequences. However, these sequences are currently described using 29 functional labels arranged within a hierarchy which is different to the hierarchical organisation of sequences within the Protvec embedding. Furthermore, by representing unknown sequences with Protvec, I demonstrated that unknown sequences form clusters that likely share related biological roles. Such clusters may be beneficial for selecting optimal candidate proteins to characterize experimentally.
Doi: https://doi.org/10.1186/s12859-022-04930-5External link
Comparison of Bacillus carbohydrate metabolism sequences grouped using agglomerative clustering on sequence embeddings using the Bacillus carbohydrate metabolism Protvec model and the SEED annotation hierarchy. The color joining the dendrograms is continuous across the Protvec dendrogram. Boxes are drawn around each subsystem in the SEED annotation hierarchy.
Illustration: Susie Grigson